# NVlabs/Sana

SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer

Repository: https://github.com/NVlabs/Sana
Canonical: https://ross.abutalabs.com/products/sana
Homepage: https://nvlabs.github.io/Sana/docs/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: diffusion, dit, pytorch, sana, text-to-image-generation, transformers, video-generation, linear-transformer, nvfp4, reinforcement-learning, system-algorithm-deisgn, text-to-video, world-models, streaming-video
Last push: 2026-08-26T14:02:08+00:00

## Health v2 (maintenance only)
Score: 74/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 55, longevity 49
- inputs: {"age_days": 691, "days_push": 7, "days_rel": 85, "gap_med": 441, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 8833, forks 710 (observed 2026-08-28T04:10:25.920966+00:00)

## What it is
SANA is an efficiency-oriented PyTorch codebase for high-resolution text-to-image and text-to-video generation built on Linear Diffusion Transformers. It provides complete training and inference pipelines for models including Sana, Sana-Sprint (few-step distillation), Sana-Video, and LongSana real-time video generation.

## Use cases
- generate 4K images from text prompts
- efficient text-to-image generation on consumer GPUs with limited VRAM
- one-step or few-step image generation with distilled models
- generate 720p video from text or image prompts
- train or fine-tune diffusion transformer models
- run real-time streaming video generation
- post-train image and video models with reinforcement learning

## When to choose
- you need high-resolution image or video generation with limited compute or VRAM
- you want fast few-step inference via consistency distillation
- you need a full open training and inference pipeline for diffusion transformers
- you want to experiment with linear attention for generative models

## When to avoid
- you need a simple plug-and-play image generator without GPU setup
- you require non-Apache-licensed or fully closed model weights
- your project depends on standard softmax attention architectures incompatible with linear DiT

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, image-processing, video-processing, llm-training
- domain: deep-learning, image-processing, artificial-intelligence, gpu-computing
- platform: python
- tags: text-to-image, text-to-video, diffusion-transformer, linear-attention, consistency-distillation, pytorch, high-resolution-generation, world-models, video, gpu, linux, docker

## Member repositories
- NVlabs/Sana (main) score 74

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:25.920966+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:24:49.428682+00:00, confidence not recorded.
  - readme: https://github.com/NVlabs/Sana (fetched 2026-08-28T04:10:25.920966+00:00, sha 0c52923c0898)
  - homepage: https://nvlabs.github.io/Sana/docs/ (fetched 2026-08-29T08:25:12.395198+00:00, sha 23183fde430f)
- Data as of 2026-08-30T08:39:29.467469+00:00.